Content uploaded by Iftikhar Ahmad
Author content
All content in this area was uploaded by Iftikhar Ahmad on Feb 13, 2020
Content may be subject to copyright.
Content uploaded by Iftikhar Ahmad
Author content
All content in this area was uploaded by Iftikhar Ahmad on Jul 23, 2019
Content may be subject to copyright.
UNCORRECTED PROOF
ARTICLE INFO
Article history:
Received 20 November 2015
Received in revised form 22 May 2016
Accepted 25 May 2016
Available online xxx
Keywords:
Raman spectroscopy
Ovarian cancer
Support vector machine classifier
Computer assisted classification
ABSTRACT
Conventional screening tools for ovarian cancer such as cancer antigen (CA-125) and trans-pelvic ultrasound have poor
sensitivity and specificity, indicating the need for better and more reliable screening methodologies. Here, we investi-
gate the capability of Raman spectroscopy as a screening technique for ovarian cancer. Raman spectra from the blood
serum of healthy control and ovarian cancer subjects were measured. Highly significant Raman shifts (p < 0.0001) and
intensity variations were observed in the cancer group as compared to the healthy group. These spectral differences were
exploited by support vector machine classifier towards computer assisted classification. Calculated evaluation metrics
such as sensitivity (=90), specificity (=100), positive predictive value (=100) and negative predictive value (=87.5) for
such classification indicated that these results are promising, with potential future application of Raman spectroscopy for
ovarian cancer screening.
© 2016 Published by Elsevier Ltd.
Photodiagnosis and Photodynamic Therapy xxx (2016) xxx-xxx
Contents lists available at ScienceDirect
Photodiagnosis and Photodynamic Therapy
journal homepage: www.elsevier.com
Computer assisted optical screening of human ovarian cancer using Raman
spectroscopy
Irfan Ullah,aIftikhar Ahmad,a, ⁎Hasan Nisar,aSaranjam Khan,bRahat Ullah,bRashad Rashid,bHassan Mahmood c
aPakistan Institute of Engineering and Applied Science (PIEAS), Nilore 45650, Islamabad, Pakistan
bNational Institute of Lasers and Optronics (NILOP), Nilore 45650, Islamabad, Pakistan
cCiti Lab, Islamabad, Pakistan
1. Introduction
Ovarian adenocarcinoma, notorious for its silent lethality, is the
second leading gynecological malignancy accounting for 5% of all
women cancers [1] and is the fifth major cause of cancer related deaths
[2], indicating it’s significantly higher ratio of incidence to death [3].
Effective screening for early detection of epithelial ovarian cancer
may contribute towards timely treatment and thus substantial reduc-
tion in the morbidity and mortality rates. However, it appears that
there is presently no sufficiently accurate screening test to this end.
Typically, early epithelial ovarian cancer has no obvious symp-
toms [4]. Nevertheless, studies have indicated that the most frequent
signs of ovarian cancer include abdominal ascites, pelvic or abdom-
inal pain and pressure symptoms such as urinary urgency or fre-
quency etc. [5]. It is noteworthy that abnormal vaginal bleeding has
been considered as a symptom of uterine and cervical cancers but is
rarely observed in ovarian cancer [6]. Pelvic examination, though an
integral component of physical evaluation of all patients suspected
of having ovarian cancer, only seldom reveals any findings sugges-
tive of the disease during early stage of disease [7]. Nevertheless, a
thorough pelvic examination in combination with blood test for the
quantification of cancer antigen (CA-125) concentrations and an ab-
domino-pelvic ultrasound may be offered to evaluate women with
symptoms or at high risk. This strategy, however, has not proven ef-
fective for early cancer detection, probably due to significantly lower
sensitivity of CA-125. A recent clinical trial in average-risk women
illustrated that these tests (used as screening tool) had no impact on
⁎⁎ Corresponding author.
Email address: iahmadmp@gmail.com (I. Ahmad)
ovarian cancer mortality [2]. Some studies indicated that CA-125 in
combination with other tumor markers e.g. D-dimer can improve the
sensitivity [8,9], but these are not used routinely and may prove to be
expensive, time consuming and cumbersome. There is hence a need
for improved screening strategies.
Due to the relatively non-invasive nature of blood sampling, the
CA-125 assay has been used as a first-line screening for ovarian can-
cer. However, a large range of malignant diseases (uterine, fallop-
ian tube, pancreas, stomach, colon, rectum cancers) [10] and non-ma-
lignant conditions (benign ovarian tumor e.g. Meigs syndrome, en-
dometriosis, salpingitis, pelvic inflammatory disease, pregnancy,
menstruation, leiomyoma, diverticulosis, pancreatitis, etc.) [11] are
also known to elevate CA-125 levels. Further, metastases from other
sites including breast and lung can also elevate CA-125 levels. Due
to poor sensitivity and specificity, it is not practical to use CA-125
alone for screening and early detection of ovarian cancer. Although,
CA-125 levels are elevated in 23–50% of stage I and in 90% stage II
ovarian cancer patients [1]; however, the marker alone has not proven
to be sufficiently effective in screening [12]. Consequently, additional
or complementary tests are essentially needed to make CA-125 a use-
ful component of a screening program. That said, the most important
sign is the presence of a pelvic mass on recto- vaginal examination.
Abdomino-pelvic ultrasound can potentially add to the assessment of
such pelvic mass (overall sensitivity of 84.9%) [13]. A solid, irregu-
lar, fixed pelvic mass is highly suggestive of an ovarian malignancy.
However, it is hard to differentiate between ovarian cancer and other
common conditions such as ovarian cyst and endometriosis on ultra-
sound, particularly if the ovarian volume is normal. Also, its use for
screening is limited by the cost of annual screening in a general pop-
ulation. The final diagnosis ultimately requires an exploratory laparo-
tomy.
http://dx.doi.org/10.1016/j.pdpdt.2016.05.011
1572-1000/© 2016 Published by Elsevier Ltd.
UNCORRECTED PROOF
2 Photodiagnosis and Photodynamic Therapy xxx (2016) xxx-xxx
A variety of optical imaging and diagnostic techniques such as
photoacoustic imaging (PAI) [14], optical coherence tomography
(OCT) [15], nonlinear microscopy (NLM) [16,17], Fourier transform
infrared spectroscopy (FTIR) [18], etc. have been investigated for
screening and early detection of ovarian cancer with each having its
own advantages and shortcomings. Specifically, PAI benefits from a
large penetration depth (∼1 cm in tissue) and depth-resolved images,
enabling the in vivo evaluation. However, potential issues that limit
PAI application would include high acoustic attenuation and relatively
complex instrumentation and signal processing. OCT has superior res-
olution (axially and laterally) in turbid media, and also provides depth-
resolved images thru the mechanism of coherence gating. However, it
offers reduced field of view and again requires more complicated op-
tics. NLM allows assessment of unfixed, unsectioned, and unstained
tissues at high resolution (comparable to histologic sections) facili-
tating the detection of very early cellular changes in the ovarian sur-
face epithelium. Specifically, a red shift in intrinsic fluorescence and
collagen structural alterations has been identified as cancer-associ-
ated changes. However, it offers small field of view and shallow sam-
pling depth. FTIR is a relatively simple and rapid technique with a
minimal sample preparation and can be used for both qualitative and
quantitative analysis of various biochemical components in a sam-
ple [18]. However, FTIR spectroscopy has very limited spatial resolu-
tion and the sample spectrum limits to reflect the average biochemi-
cal information referred to the whole probed sample. In addition, Ra-
man spectroscopy has attracted great interest for detection of many
cancers [19,20] including ovarian cancer [21]; the mean Raman spec-
trum from ovarian cancer tissues exhibited a broader amide I band, a
stronger amide III band, a minor blue shift in the δCH2band (δrep-
resent stretching deformation), and a hump around 1480 cm−1 com-
pared to the spectrum of normal ovarian tissues, indicating that Ra-
man spectroscopy can sensitively measure the variations in molecu-
lar chemistry of ovarian tissue samples potentially enabling assess-
ment of early cancer changes. Nevertheless, Raman spectroscopy is
relatively unexplored for screening/early detection of ovarian cancer.
Due to the relatively simple, rapid and non-invasive nature of blood
sampling, Raman spectroscopy for screening of ovarian cancer would
comprise an interesting study of clinical importance. That said, Owens
et al. studied Raman (and FTIR) spectroscopy coupled with multivari-
ate analysis for blood plasma samples towards early detection of ovar-
ian cancer and concluded that this approach facilitates the identifica-
tion of spectral alterations associated with the presence of ovarian can-
cer [22].
In this study, we interrogate the capability of Raman spectroscopy
to assess blood samples from ovarian cancer patients for possible opti-
cal signatures towards screening and early detection. Specifically, we
investigate possible differences in Raman spectra collected from blood
samples of healthy and ovarian cancer subjects. Such spectroscopic
changes may be correlated to the underlying biochemical changes.
Furthermore, computer assisted classification was used to aid the dis-
crimination between the two groups, based on the Raman spectral sig-
natures. Raman spectroscopy in tandem with computer aided classifi-
cation algorithm may provide a simple, fast and inexpensive tool for
screening of ovarian cancer. It may also help to monitor treated cases
of ovarian cancer for disease recurrence.
2. Materials and methods
2.1. Subjects and protocol
Blood samples from 11 patients with confirmed clinical and
histopathological ovarian cancer and 11 healthy volunteers that
matched the case group in demographic profile including median age,
race and gender were used to study the possible spectroscopic signa-
tures of ovarian cancer compared to healthy samples. Age characteris-
tics for control and cancer groups were: (min, max, median age) = (40,
65, 55 years) and (44, 65, 57.5 years), respectively. Blood chemistry
including CA-125 was also evaluated for both groups. The inclusion
criteria for healthy control group volunteers comprised of: normal
menstrual history, if premenopausal then currently not having menses,
normal abdomino-pelvic ultrasound and normal serum βhCG.
2.2. Serum preparation
The blood collected from the subjects was stored in red topped
tubes available from Becton Dickinson (BD) for 45 min during which
period the clotting occurred. Afterwards, all sera samples were ex-
tracted with the help of blood centrifugation (2500 rpm for 20 min).
The separated sera samples were transferred into polypropylene tubes
using Pasteur pipettes and then stored at −20 °C till final Raman spec-
troscopic measurement.
2.3. Raman spectroscopy analysis
Raman spectra from each serum sample were measured using Ra-
man spectrometer (Dongwoo Optron, South Korea). Fig. 1 shows the
white light photo of the experimental setup. The light beam for prob-
ing the samples was obtained from the second harmonic of diode
pumped Nd-YAG laser (wavelength 532 nm). A 100X objective lens
was used for dual purpose: to properly direct the incident light on the
sample; and to focus the light after interaction on the detector in back
scattering configuration. Raman spectra were acquired for each sam-
ple in the spectral range of 500–2000 cm−1 with a spectral resolution
and acquisition time of 4 cm−1 and 10 s, respectively.
2.4. Data preprocessing
A total of 42 Raman spectra (total samples 21; each sample ana-
lyzed twice) were collected. We used Savitzky–Golay (SG) filter for
smoothing the measured Raman spectra which improved the signal to
noise ratio (SNR) while preserving the integrity of inherently weak
Raman peaks [20]. The fluorescence contribution towards the Raman
spectra was removed by the cubic spline interpolation method fol-
lowed by spectra normalization. Mean Raman spectra for both groups
were calculated from all individual spectra in each group (n = 11) and
subsequently compared. All data processing was performed in Matlab.
2.5. Statistical analysis
The use of support vector machine (SVM) classifiers in the arena
of medical diagnosis and classification is attracting increasing inter-
est, as they have the capability for accurate classification in signifi-
cantly shorter time and without subjectiveness [23,24]. Once trained,
SVM functions could potentially classify input patterns. For better un-
derstanding, the implementation of SVM algorithm can be divided
into three major steps. First, the input data set with non-linear behav-
ior is mapped into higher-dimensional feature space where the indi-
vidual characteristics of the data are separable and linear classifica-
tion is thereby possible. Second, linear classification in the higher-
dimensional feature space is executed. Third, the data set is trans-
formed back to the original nonlinear space. SVMs make use of spe
UNCORRECTED PROOF
Photodiagnosis and Photodynamic Therapy xxx (2016) xxx-xxx 3
Fig. 1. Illustration of the workflow of the study; (a) sample extraction and serum preparation, (b) Raman spectra acquisition and preprocessing, (c) Raman spectra and (d) support
vector machine (SVM) classification of the samples. TP = true positive; TN = true negative; FP = false positive; FN = false negative. (For interpretation of the references to colour in
the text, the reader is referred to the web version of this article.)
cialized nonlinear kernel functions for the transformation of data be-
tween the two spaces.
We developed a custom SVM algorithm towards computer assisted
classification of healthy and ovarian cancer samples based on the dif-
ferences in Raman spectra. First, the differences in the peak positions
of the two groups were statistically evaluated by calculating their p
values from unpaired two tailed t-test: all peaks of Raman spectra with
significant differences were grouped in three categories for classifica-
tion purpose, p < 0.05 (five peaks), p < 0.01 (one peak) and p < 0.0001
(six peaks). In the second step, these categories individually and their
combination were used in SVM analyses for sample classification.
Specifically, four samples each from healthy and cancer group were
used to train SVM algorithm for blind classification of the remaining
samples. Finally, the performance of the algorithm was assessed with
the help of evaluation metrics such as sensitivity, specificity, positive
predictive values (PPV) and negative predictive values (NPV) accu-
racy. All data manipulation was performed in Matlab.
3. Results
The mean Raman spectra for both healthy control (blue) and ovar-
ian cancer (red) groups have been shown in Fig. 1. To elaborate the
differences between the mean spectra of the two groups, magnified
view of few representative Raman peaks and their shifts is also given
in Fig. 1. The statistical significance of these differences was deter-
mined by two tailed unpaired t-test. A quantitative comparison of the
peaks positions, their assignment to various biomolecules and the
level of significance in their differences (p-values) are depicted in
Table 1.
The observed differences in the Raman spectra of the two sam-
ple groups can be divided in two categories; differences in amplitude
(intensity) of Raman peaks and differences in peak positions. For the
Table 1
Comparison of peak positions for healthy control and ovarian cancer groups, the signif-
icance in their differences (p-values) and their assignment to biomolecules [25].
Peak position p-value Peak assignment
Healthy
group
Cancer
group
653 641 <0.05 C C twisting mode of tyrosine
751 749 <0.05 Symmetric breathing of tryptophan
846 859 <0.01 Ring breathing mode of tyrosine and C C stretch
of proline ring
952 950 <0.05 Hydroxyapatite/carotenoid/cholesterol
1001 1003 <0.05 Symmetric ring breathing mode of phenylalanine
1272 1277 <0.0001 Amide III: a-helix
1326 1323 <0.0001 CH3CH2 wagging mode of collagen
1443 1447 <0.0001 CH2 bending mode of proteins, lipids, fatty acids
1508 1511 <0.05 C C carotenoid/N H bending
1598 1597 <0.0001 C C in-plane bending mode of phenylalanine and
tyrosine
1655 1657 <0.0001 Amide I (C O stretching mode of proteins, a-
helix conformation)/C C lipid stretch
1740 1744 <0.0001 C O stretch (lipid)
UNCORRECTED PROOF
4 Photodiagnosis and Photodynamic Therapy xxx (2016) xxx-xxx
amplitude differences, the CH2peak at 1447 cm−1, the amide I peak
at 1657 cm−1 and the C O stretching peak at 1744 cm−1 showed
maximum differences. These peaks were assigned to the bending of
proteins/lipids/fatty acids, the amide I stretching of protein backbone
and the stretching of lipids [25,26], respectively. Minimum ampli-
tude differences were observed for peaks at 641 cm−1, 749 cm−1 and
950 cm−1. These peaks were allotted to the stretching of C–S in cys-
teine, the symmetric breathing of tryptophan and the hydroxyapatite/
carotenoid/cholesterol [25], respectively. The peaks with intermedi-
ate amplitude differences were 859 cm−1, 1003 cm−1 and 1323 cm−1.
These peaks were assigned to the ring breathing of tyrosine, the sym-
metric ring breathing of phenylalanine and the wagging mode of
CH3CH2, respectively [25,27]. Further, the slightly visible shoulders
at 1277 and 1597 cm−1 in the healthy group became very prominent
the cancer group; a shoulder appears at the blue side of 1150 cm−1
peak in the cancer group. The peak values mentioned here (and in Fig.
2) represents the cancer group; the corresponding peak values for nor-
mal group are given in Table 1.
The observed differences between peak positions for the two sam-
ple groups and their statistical assessment with unpaired two
tailed t-test are summarized in Table 1. Based on the level of statisti-
cal significance (p-value), the peak position differences can be divided
in three groups; six peaks were covered in the group with p < 0.0001
while five peaks in the p < 0.05 group. The 846 cm−1 peak was the
only member of p < 0.01 group.
SVM classifier was used to assess the discrimination power of Ra-
man spectroscopy for healthy and cancerous ovarian serum samples;
the classification evaluation metrics are presented in Table 2. SVM
was individually applied to Raman spectral peaks with a) p < 0.05,
b) p < 0.0001 and c) their combination. SVM classified the samples
with sensitivity (Sn = 87.5), specificity (Sp = 100), positive predictive
value (PPV = 100) and negative predictive value (NPV = 87.5) based
on the peaks having p < 0.0001. Results of SVM classification based
on peaks with p < 0.05 indicated relatively lower Sp = 85.7 and
PPV = 87.5. Further, the SVM classification results were improved
when combination of all spectral peaks (p < 0.05, p < 0.01 and
p < 0.0001) was used, as indicated by the evaluation parameters
shown in Table 2.
4. Discussion
In this study, we investigated Raman spectroscopy for the quantita-
tive assessment and comparison of blood serum samples from healthy
control and ovarian cancer subjects. SVM classifier was assessed for
automated classification of the samples.
The observed differences in the mean Raman spectra of the two
sample groups suggest the differential expression of various proteins
as well as possible changes in their conformation and composition in
the cancer group. Indeed, a large number of proteins (∼160) [28] are
overexpressed in ovarian cancer; the most common of these include
CA-125, human epididymis protein 4 (HE4) [29], haptoglobin [30],
osteopontin [31], mesothelin [32], etc. This overexpression of vari-
ous proteins in ovarian cancer may be responsible for the higher peak
amplitudes in the Raman spectra. In addition, the Raman peak shifts
could be associated with the conformational changes of these and
other related biomolecules. For instance, the amide I peak (1657 cm−1)
is directly related to the backbone conformation of proteins. Thereby,
the increase in amplitude and shift in position of this peak as seen
for cancer group indicate the underlying up-regulation and structural
modifications of proteins resulting from ovarian cancer. Variations
(both in amplitude and position) in other Raman peaks for cancer
group such as amide III, 859, 1003 and 1597 cm−1 peaks suggest ad-
ditional quantitative and conformational change in various proteins
[26,33]. In addition to changes in proteins, possible alterations in the
structure and quantity of lipids are suggested by changes in 1447,
1657 and 1744 cm−1 peaks. The first two peaks show an in
Fig. 2. Mean Raman spectra from blood sera of ovarian cancer (red) and control (blue) groups. The mean spectra was obtained by averaging all samples (n = 11) in each group. The
inset shows magnified view of few representative Raman peaks and their shifts between ovarian cancer and control groups. (For interpretation of the references to colour in this figure
legend, the reader is referred to the web version of this article.)
UNCORRECTED PROOF
Photodiagnosis and Photodynamic Therapy xxx (2016) xxx-xxx 5
Table 2
Evaluation parameters of SVM classification based on differences in Raman spectra of
healthy and cancerous ovarian samples.
Raman shifts with p value Sn Sp PPV NPV
<0.0001 87.5 100 100 87.5
<0.01 100 72.7 66.7 100
<0.05 87.5 85.7 87.5 85.7
<0.0001 + < 0.01 + < 0.05 90 100 100 87.5
crease in amplitude that may reflect greater synthesis of lipids, pro-
teins and lipoproteins in ovarian cancer which is not surprising as ma-
lignant epithelial cells are known for their rapid metabolism that may
transiently increase content of such metabolites in blood due to cell
membrane breakdown. However, the third peak though more specific
for lipids according to Table 1, was found to be negative in our study.
We are unable to provide an objective explanation for this phenom-
enon and this question is open for further introspection and investi-
gations. The shoulder on the blue side of 1447 cm−1 peak indicates
alterations in the secondary structures of lipids [21]. These Raman
biomarkers should be further investigated for possible utilization in
the screening of ovarian cancer either alone or in adjunct with other
screening tools. The amide I peak could be a better candidate for such
discrimination as it is placed outside the more crowded fingerprint re-
gion and has better signal intensity.
Based on the Raman spectral signatures, SVM algorithm towards
automated classification illustrated encouraging results. Although
biopsy is the gold standard and the only acceptable means for diag-
nosis in almost all cancers, the NPV of 100% in our results may lead
to further studies attempting to highlight a subset of patients with sus-
pected ovarian cancer who can be spared from invasive biopsy pro-
cedures and put on close monitoring or follow-up instead, an option
which is currently a risky one with CA-125 screening due to low NPV
of CA-125. That said, the serum samples analyzed in this study were
collected from late stage ovarian cancer patients; thereby the observed
Raman changes in this study may not be inevitably present at early
stages of ovarian cancer. To address this limitation, Raman studies
with early stage ovarian cancer sera samples are suggested.
A comparison of this study with present clinical practice and with
optical techniques in research phase has been shown in Table 3; the
comparison indicates that Raman spectroscopy appears to have better
efficiency for ovarian cancer screening than other contemporary op-
tical techniques such as photoacoustic imaging (PAI), optical coher-
ence tomography (OCT), reflectance spectroscopy and fluorescence
microscopy in terms of Sn,Sp,PPV and NPV. Moreover, our find-
ings reveal that Raman spectroscopy may have comparable efficiency
to polarization sensitive optical coherence tomography (PS-OCT) for
ovarian cancer detection. However, studies with much larger sample
cohort and from patients of various age groups (such as pre- and post-
menopause) are required for more robustly validated Raman spectral
Table 3
Evaluation parameters of classification for cancerous ovarian samples based on various
techniques.
Screening technique Sn Sp PPV NPV Ref
CA-125 50–62 95 57 70.6 [36]
Ultrasound 85 98 5.3 [13]
PAI 83 83 [14]
Fluorescence microscopy 88 93 [37]
OCT 75 80 [38]
PS-OCT 100 83 80 100 [39]
Reflectance spectroscopy 86 80 [40]
Raman spectroscopy 90 100 100 87.5 This study
analyses. Indeed, we have begun working on such a comprehensive
Raman spectral study; determination of spectral finger prints for both
tissue and blood samples validated by histology. For completeness,
a comprehensive list of markers for ovarian cancer screening (in re-
search phase) can be found in [34,35].
One particularly interesting study of Raman spectroscopy for com-
parison has been recently reported by Owens et al. where Raman
and FTIR spectra analyses of blood plasma followed by SVM clas-
sifier showed a diagnostic accuracy of 74% and 93.3%, respectively
[22]. Specifically, Owens et al. mainly focused on FTIR spectroscopy
for the detection of ovarian carcinoma using blood serum samples
whereas Raman spectroscopy was used for only selected number of
blood plasma samples. On the other hand, the present study focuses
purely on Raman spectroscopy for screening ovarian carcinoma us-
ing blood serum samples. Using serum rather than plasma may have
significant impact on sample analysis. For instance, plasma samples
require sampling tubes pre-treated with anti-coagulant (which is not
required for serum samples). In particular, the anticoagulant quantity
and protocol varies from one laboratory to another. Anticoagulants
like heparin pose a risk of being contaminated that may even stimu-
late white blood cells to release cytokines into the sample. Further-
more, lipemic, icteric or hemolysed plasma samples often result in
ambiguities during analysis. Contrarily, serum samples do not require
the use of anticoagulant during preparation resulting in elimination of
the aforementioned problems which may add to overall reliability and
consistency of results. Nevertheless, the findings of these two studies
(Owens et al. and present study) support each other which might stim-
ulate the interest of other relevant research groups in further analyzing
the utility of Raman Spectroscopy towards optical screening of ovar-
ian cancers ultimately benefiting the patients.
5. Conclusion
This study illustrates the capability of Raman spectroscopy for the
assessment of ovarian cancer. Specifically, changes in peak positions
and intensity of Raman spectra for cancer serum samples as compared
to healthy samples were successfully quantified. These spectral signa-
tures presumably indicate structural alterations in pre-existing biomol-
ecules and presence of new molecules. SVM classifier was used to dis-
criminate these observed differences towards computer aided sample
classification/screening. The automated classification efficiency was
evaluated by metrics such as sensitivity, specificity, positive and neg-
ative predictive values. The findings of this study indicate that spectral
changes caused by ovarian cancer can be sensitively measured with
Raman spectroscopy, with potential future application for ovarian can-
cer screening in tandem with CA-125.
Acknowledgements
Citi Labs, Islamabad and Rawalpindi, Pakistan are acknowledged
for providing the cancerous serum samples. The authors are grateful
to Dr. S. M. Mirza for his valuable suggestion regarding classification
algorithms.
References
[1] N. Colombo, et al., Ovarian cancer, Crit. Rev. Oncol Hematol. 60 (2) (2006)
159–179.
[2] Cancer facts and Figs 2015, Am. Cancer Soc. 18 (2015).
[3] R.L. Siegel, K.D. Miller, A. Jemal, Cancer statistics, CA Cancer J. Clin. 65 (1)
(2015) 5–29.
[4] S.E. Taylor, J.M. Kirwan, Ovarian cancer: current management and future di-
rections, Obstet. Gynaecol. Reprod. Med. 22 (2) (2012) 33–37.
UNCORRECTED PROOF
6 Photodiagnosis and Photodynamic Therapy xxx (2016) xxx-xxx
[5] A.E. Guppy, P.D. Nathan, G.J.S. Rustin, Epithelial ovarian cancer: a review of
current management, Clin. Oncol. 17 (6) (2005) 399–411.
[6] L. Gilbert, et al., Assessment of symptomatic women for early diagnosis of
ovarian cancer: results from the prospective DOvE pilot project, Lancet On-
col. 13 (3) (2012) 285–291.
[7] J.D. Seidman, P. Zhao, A. Yemelyanova, ‘Primary peritoneal’high-grade
serous carcinoma is very likely metastatic from serous tubal intraepithelial car-
cinoma: assessing the new paradigm of ovarian and pelvic serous carcinogene-
sis and its implications for screening for ovarian cancer, Gynecol. On-
col. 120 (3) (2011) 470–473.
[8] A. Gadducci, et al., Preoperative evaluation of D-dimer and CA 125 levels in
differentiating benign from malignant ovarian masses, Gynecol. Oncol. 60 (2)
(1996) 197–202.
[9] R.P. Woolas, et al., Combinations of multiple serum markers are superior to in-
dividual assays for discriminating malignant from benign pelvic masses, Gy-
necol. Oncol. 59 (1) (1995) 111–116.
[10] E.L. Moss, et al., Views of general practitioners on the role of CA125 in pri-
mary care to diagnose ovarian cancer, BMC Womens Health 13 (8) (2013) 2–7.
[11] U. Menon, M. Grif, A. Gentry-maharaj, Ovarian cancer screening–current sta-
tus, future directions, Gynecol. Oncol. 132 (2) (2014) 490–495.
[12] I.J. Jacobs, U. Menon, Progress and challenges in screening for early detection
of ovarian cancer, Mol. Cell. Proteomics 3 (4) (2004) 355–366.
[13] U. Menon, et al., Sensitivity and specificity of multimodal and ultrasound
screening for ovarian cancer, and stage distribution of detected cancers: results
of the prevalence screen of the UK Collaborative trial of ovarian cancer screen-
ing (UKCTOCS), Lancet Oncol. 10 (2009) 327–340.
[14] A. Aguirre, et al., Potential role of coregistered photoacoustic and ultrasound
imaging in ovarian cancer detection and characterization, Transl. Oncol. 4 (1)
(2011) 29–37.
[15] T. Wang, M. Brewer, Q. Zhu, An overview of optical coherence tomography
for ovarian tissue imaging and characterization, WIREs: Nanomed.
Nanobiotechnol. 7 (1) (2015) 1–16.
[16] R.M. Williams, et al., Strategies for high resolution imaging of epithelial ovar-
ian cancer by laparoscopic nonlinear microscopy, Transl. Oncol. 3 (3) (2010)
181–194.
[17] J. Adur, et al., Optical biomarkers of serous and mucinous human ovarian tumor
assessed with nonlinear optics microscopies, PLoS One 7 (10) (2012) 1–13.
[18] R. Mehrotra, et al., Analysis of ovarian tumor pathology by fourier transform
infrared spectroscopy, J. Ovarian Res. 3 (1) (2010) 27.
[19] K. Kong, et al., Raman spectroscopy for medical diagnostics—from in-vitro
biofluid assays to in-vivo cancer detection, Adv. Drug Deliv. Rev. 89 (2015)
121–134.
[20] W. Wang, et al., Real-time in vivo cancer diagnosis using raman spectroscopy,
J. Biophotonics 8 (7) (2015) 527–545.
[21] K. Maheedhar, et al., Diagnosis of ovarian cancer by Raman spectroscopy: a pi-
lot study, Photomed. Laser Surg. 26 (2) (2008) 83–90.
[22] G.L. Owens, et al., Vibrational biospectroscopy coupled with multivariate
analysis extracts potentially diagnostic features in blood plasma/serum of ovar-
ian cancer patients, J. Biophotonics 209 (3) (2014) 200–209.
[23] H. Wang, G. Huang, Application of support vector machine in cancer diagno-
sis, Med. Oncol. 28 (1) (2011) 613–618.
[24] N.H. Sweilam, a.a. Tharwat, N.K. Abdel Moniem, Support vector machine for
diagnosis cancer disease: a comparative study, Egypt. Inform. J. 11 (2) (2010)
81–92.
[25] N. Stone, et al., Raman spectroscopy for identification of epithelial cancers,
Faraday Discuss. 126 (2004) 141–157.
[26] A.L. Jenkins, R. a. Larsen, T.B. Williams, Characterization of amino acids us-
ing Raman spectroscopy, Spectrochim. Acta Part A Mol. Biomol. Spec-
trosc. 61 (7) (2005) 1585–1594.
[27] K.W. Short, et al., Raman spectroscopy detects biochemical changes due to pro-
liferation in mammalian cell cultures, Biophys. J. 88 (6) (2005) 4274–4288.
[28] B.M. Nolen, A.E. Lokshin, Protein biomarkers of ovarian cancer: the forest and
the trees, Future Oncol. 8 (1) (2012) 55–71.
[29] J. Li, et al., HE4 as a biomarker for ovarian and endometrial cancer manage-
ment, Expert Rev. Mol. Diagn. 9 (6) (2009) 555–566.
[30] C. Zhao, et al., Circulating haptoglobin is an independent prognostic factor in
the sera of patients with epithelial ovarian cancer, Neoplasia 9 (1) (2007) 1–7.
[31] J.O. Schorge, et al., Osteopontin as an adjunct to CA125 in detecting recurrent
ovarian cancer, Clin. Cancer Res. 10 (2004) 3474–3478.
[32] R. Hassan, et al., Detection and quantitation of serum mesothelin, a tumor
marker for patients with mesothelioma and ovarian cancer, Clin. Cancer
Res. 12 (2) (2006) 447–454.
[33] Z.-Q. Wen, Raman spectroscopy of protein pharmaceuticals, J. Pharm.
Sci. 96 (11) (2007) 2861–2878.
[34] E. Kobayashi, et al., Biomarkers for screening, diagnosis, and monitoring of
ovarian cancer, Cancer Epidemiol. Biomarkers Prev. 21 (11) (2012) 1902–1912.
[35] K. Eagle, J. Ledermann, Tumor markers in ovarian malignancies, Oncolo-
gist 2 (5) (1997) 324–329.
[36] V. Nossov, et al., The early detection of ovarian cancer: from traditional meth-
ods to proteomics. Can we really do better than serum CA-125?, Am. J. Obstet.
Gynecol. 199 (3) (2008) 215–223.
[37] R. George, et al., Parallel factor analysis of ovarian autofluorescence as a cancer
diagnostic, Lasers Surg. Med. 44 (4) (2012) 282–295.
[38] Y. Yang, et al., Optical scattering coefficient estimated by optical coherence to-
mography correlates with collagen content in ovarian tissue, J. Biomed.
Opt. 16 (9) (2011) 090504.
[39] T. Wang, Y. Yang, Q. Zhu, A three-parameter logistic model to characterize
ovarian tissue using polarization-sensitive optical coherence tomography, Bio-
med Opt. Express 4 (5) (2013) 772–777.
[40] U. Utzinger, et al., Reflectance spectroscopy for in vivo characterization of
ovarian tissue, Lasers Surg. Med. 28 (1) (2001) 56–66.
